Skip to main content
Workflows in Hyrex allow you to orchestrate multiple tasks into complex directed acyclic graphs (DAGs). Tasks can run in sequence or parallel, with automatic dependency management.

​
Basic Workflow

Define workflows using the @hy.workflow decorator and the >> operator: 
from hyrex import HyrexRegistry

hy = HyrexRegistry()

@hy.task
def extract():
    # Extract data from source
    data = fetch_from_database()
    return {"count": len(data), "data": data}

@hy.task
def transform():
    # Transform the data
    processed = apply_transformations()
    return {"processed_count": len(processed)}

@hy.task
def load():
    # Load to destination
    write_to_destination()
    return {"status": "completed"}

@hy.workflow
def etl_workflow():
    # Define task dependencies
    extract >> transform >> load

​
Parallel Execution

Run multiple tasks in parallel using list notation: 
@hy.task
def fetch_source_a():
    return {"source": "a", "data": get_data_a()}

@hy.task 
def fetch_source_b():
    return {"source": "b", "data": get_data_b()}

@hy.task
def fetch_source_c():
    return {"source": "c", "data": get_data_c()}

@hy.task
def combine_sources():
    # This runs after all parallel tasks complete
    return {"combined": True}

@hy.workflow
def parallel_workflow():
    # Run fetch tasks in parallel, then combine
    [fetch_source_a, fetch_source_b, fetch_source_c] >> combine_sources

​
Triggering Workflows

Use the .send() method to trigger workflow execution: 
# Simple workflow without arguments
@hy.workflow
def simple_workflow():
    extract >> transform >> load

# Trigger the workflow
workflow_run = simple_workflow.send()
For workflows with arguments, pass the context object to .send(): 
@hy.workflow(workflow_arg_schema=ETLConfig)
def parameterized_workflow():
    extract >> transform >> load

# Trigger with typed arguments
workflow_run = parameterized_workflow.send(ETLConfig(
    source_table="events",
    destination_table="processed_events"
))

​
Workflow Arguments

Pass typed arguments to workflows using Pydantic models: 
from pydantic import BaseModel
from typing import List

class ETLConfig(BaseModel):
    source_table: str
    destination_table: str
    filters: List[str] = []
    batch_size: int = 1000

@hy.task
def extract_with_config():
    # Access workflow args in tasks
    ctx = get_hyrex_workflow_context()
    if ctx and ctx.workflow_args:
        config = ETLConfig(**ctx.workflow_args)
        print(f"Extracting from {config.source_table}")
        # Use config.filters, config.batch_size, etc.
    return {"extracted": True}

@hy.workflow(workflow_arg_schema=ETLConfig)
def parameterized_etl():
    extract_with_config >> transform >> load

# Send workflow with arguments
workflow_run = parameterized_etl.send(ETLConfig(
    source_table="raw_events",
    destination_table="processed_events",
    filters=["status = 'active'"],
    batch_size=5000
))

​
Access Workflow Context

Tasks within a workflow can access the workflow context and other task results: 
from hyrex import get_hyrex_workflow_context

@hy.task
def validate_extraction():
    context = get_hyrex_workflow_context()
    
    if context:
        # Access workflow arguments
        if context.workflow_args:
            config = ETLConfig(**context.workflow_args)
            print(f"Validating extraction for {config.source_table}")
        
        # Access other task results by task name
        extract_run = context.durable_runs.get("extract_with_config")
        if extract_run:
            # Refresh to get latest status
            extract_run.refresh()
            
            # Check task runs
            for run in extract_run.task_runs:
                if run.status == "completed":
                    print(f"Extract result: {run.result}")
    
    return {"validation": "passed"}

​
Task Configuration in Workflows

Override task configuration within workflows: 
@hy.workflow(
    queue="etl-workflows",  # Default queue for all tasks
    priority=7
)
def configured_workflow():
    # Extract uses workflow defaults
    extract >> \
    # Transform needs more resources
    transform.with_config(
        queue="cpu-intensive",
        max_retries=5,
        timeout_seconds=600
    ) >> \
    # Load uses different queue
    load.with_config(
        queue="io-heavy",
        max_retries=3
    )

​
Complex Workflow Patterns

Build sophisticated DAGs with multiple parallel and sequential stages: 
@hy.workflow
def complex_etl():
    # Stage 1: Extract from multiple sources in parallel
    sources = [extract_postgres, extract_s3, extract_api]
    
    # Stage 2: Validate each source (parallel)
    validations = [validate_postgres, validate_s3, validate_api]
    
    # Connect sources to their validators
    extract_postgres >> validate_postgres
    extract_s3 >> validate_s3  
    extract_api >> validate_api
    
    # Stage 3: Merge all validated data
    validations >> merge_data
    
    # Stage 4: Transform and enrich in parallel
    merge_data >> [transform_users, transform_events, enrich_metadata]
    
    # Stage 5: Final load
    [transform_users, transform_events, enrich_metadata] >> load_warehouse

​
Next Steps

Queues

Route workflow tasks to specific workers

Monitoring

Monitor workflow execution